Using Radiomics and Convolutional Neural Networks for the Prediction of Hematoma Expansion After Intracerebral Hemorrhage

Figures & data

Figure 1 Fully automated hybrid model for HE prediction. In Model 2, we use feature-level fusion approaches for fusion of features with the AIM of collecting complementary information from radiomics, clinical data.

Table 1 The Characteristics of All Inclusion Participants

Variables	NEG-HE	POS-HE	P value
Gender (M/F)	220/90	68/30	0.739
Age (year±SD)	59.4±11.17	59.53±13.67	0.945
SBP (mmHg±SD)	170.69±27.94	172.16±30.80	0.767
DBP (mmHg±SD)	97.53±27.94	98.61±20.79	0.739
ADL Index (median)	35	5	0.0004
History of hypertension (Y/N, n)	204/106	58/40	0.109
History of heart disease (Y/N, n)	16/294	4/94	0.674
History of diabetes (Y/N, n)	18/292	6/92	1.000
Platelet count (×10^9/L, median)	179	193	0.243
Neutrophil count (×10^9/L, median)	6.84	7.79	0.288
Percentage of neutrophils (median, %)	78.7	78.1	0.788
D-dimer (mg/L, median)	0.29	0.29	0.644
Fibrinogen (g/L, median)	2.46	2.35	0.712
Prothrombin time (INR, median)	1.5	11.7	0.135
Hematoma volume (mL, median)	11.13	17.74	0.008

Abbreviations: NEG-HE, negative-hematoma enlargement; POS-HE, positive-hematoma enlargement; SBP, systolic blood pressure; DBP, diastolic blood pressure; ADL, activity of daily living scale index; Y/N, yes/no.

Figure 2 The patient enrollment process.

Figure 3 Representative image of hematoma by automatically labeled. The effect of the automatic hematoma labeling tool on a certain patient. The red color in the above picture shows the range of the labeled hematoma, and the picture below shows the original axial image of the brain.

Table 2 Results of Hybrid Module Receiver Operating Characteristic, Precision-Recall

	Precision	Recall	f1-Score	Support
NEG-HE	0.95	0.99	0.97	76
POS-HE	0.95	0.83	0.89	24
Accuracy	NA	NA	0.95	100
Macro average	0.95	0.91	0.93	100
Weighted average	0.95	0.95	0.95	100

Abbreviations: NEG-HE, negative-hematoma enlargement; POS-HE, positive-hematoma enlargement; NA, not available.

Figure 4 Performance of the hybrid model in the prediction of HE status. (A) The confusion matrix shows how well the model predicts the test set. (B) ROC curve of the model. (C) The precision-recall curve of the model. (D) Top 15 features with high importance in the SVM based classifier.